Adaptive Swarm Intelligence for Real-Time Anomaly Detection in Edge-Enabled Wireless Sensor Networks

Abstract

This thesis investigates real-time anomaly detection in resource-constrained Wireless Sensor Networks (WSNs) using an edge-based approach. A lightweight sequential detector combining Exponentially Weighted Moving Average (EWMA) and the Page-Hinkley test is deployed on ESP32 microcontrollers to ensure constant memory usage and deterministic execution. To improve detection performance, a three-tier Edge-Gateway-Fog architecture is proposed, where Particle Swarm Optimization (PSO) is applied offline at the fog layer to tune detector parameters. Experimental results based on real deployments and replay-based evaluations demonstrate stable long-term operation and efficient processing of over 1.5 million observations. The optimized model significantly reduces false positive rates and unnecessary alarm transmissions while maintaining a high detection rate. Although a slight increase in detection delay is observed, the approach provides a practical and energy-efficient solution for embedded anomaly detection in IoT environments.